Artificial neural network method for automatic mask generation for PIV: Applications in a 5x5 rod bundle with mixing vane spacer grids

Particle image velocimetry (PIV) is a non-intrusive flow visualization technique. The fluid is seeded with particles of the appropriated size and illuminated with a laser. Digital cameras capture a sequence of images allowing the tracking of the particle’s displacement, from which the velocity vectors are calculated. One of the most important advantages of PIV is its ability to capture the velocity field in a two-dimensional plane, or even a volume. However, measurements near solid interfaces are complex since the velocity vectors associated to undesirable regions, such as walls, introduce bias in the neighboring vectors. One common way to discard undesirable data is to apply a logical mask, created manually or automatically using algorithms, which defines the valid measurement regions and areas to be ignored. Some complex geometries found in nuclear reactor designs, such as pebble beds, wire-wrapped bundles and mixing vanes, are especially time-consuming or difficult to mask. They do not have regular geometries or cannot be easily described by intensity thresholds or gradients for the case of traditional automatic masking. This work presents a method for automatic masking based on feedforward artificial neural network (ANN) that identifies a pattern in the pixel intensity statistics, characterizing the areas of the image to be measured or masked. Training consists a set of binary points in which values determine whether that point is part of the mask of the measurement domain. Since the algorithm does not have any prior built-in assumption about the system, the method permits the user to identify areas of interest changing only the training points. This work presents the evaluation of this method in a PIV analysis of a Matching index-of-refraction facility of a PWR 5x5 rod bundle. The objectives are to mask the spacer grid with mixing vanes and laser glare coming from vibrating rods.